cv::ORB works awfully bad on OpenCV 3.0 Beta (Bug #4123)

Added by Erick Noe Amezquita Lucio about 10 years ago. Updated almost 10 years ago.

Status:Cancelled Start date:2015-01-15
Priority:Normal Due date:
Assignee:Maria Dimashova % Done:

0%
Category:features2d
Target version:3.0
Affected version:branch 'master' (3.0-dev) Operating System:Windows
Difficulty: HW Platform:x64
Pull request:

Description

When using

cv::Ptr<cv::FeatureDetector> featDetector;
featDetector = cv::ORB::create();

and then using

featDetector->detect and featDetector->compute, it takes quite a bit of time and after exiting the program (or thread) it creates a memory exception.

Using a feature detector like AKAZE (as in cv::AKAZE::create()) and using ->detect or ->compute works wonders, albeit it is significantly slower than ORB, at least in multithread.

This doesn't happen in OpenCV 2.4.10, albeit using

cv::Ptr<cv::FeatureDetector> featDetector;
featDetector = new cv::ORB;

cv::Ptr<cv::DescriptorExtractor> descExtractor;
descExtractor = new cv::ORB;

with featDetector->detect and descExtractor->compute, it is fast in multithread, faster than AKAZE indeed.

Thanks for your attention and this wonderfull tool, OpenCV.
Best Regards.

Associated revisions

Revision 12e6efc6
Added by Alexander Alekhin over 9 years ago

Merge pull request #4123 from rajithr:patch-1

History

#1
Updated by Vadim Pisarevsky almost 10 years ago

ORB passes our regression tests where it works reasonably good (though, worse than modern AKAZE). Please, supply the example on how you call ORB and reopen the ticket.

  • Status changed from New to Cancelled

Also available in: Atom PDF